Spatial and Temporal Analysis of Mangrove and Forest Change Based on Pre-Segmented Land Cover Data in Malang Regency

Effective monitoring of mangrove ecosystems is essential for their conservation and sustainable management. However, some regions may lack adequate remote sensing data due to factors such as temporal gaps, limited regional coverage, or suboptimal data quality, leading to reliance on available datasets. This study evaluates the Random Forest Classifier (RFC) for predicting land cover types across 16 coastal areas in Malang Regency using RGB-based land cover images. RFC achieved a high overall accuracy of up to 93% across three experiments, consistently demonstrating strong performance in identifying dominant classes, such as forest and mangroves, and outperforming the Convolutional Neural Network (CNN) model in accuracy. However, challenges were observed with minority classes, including forest loss and aquaculture ponds, reflecting the limitations of feature diversity and imbalanced data. These findings highlight the potential of RFC for analyzing land cover changes while emphasizing the importance of enhancing dataset quality and diversity to improve classification outcomes in underrepresented areas.